This is what a production line looks like that promises nothing.
This page shows what we actually computed in our own operation: how many strategies were generated and then discarded, how deep the finished portfolios' declines go, how long they last, and how alike a portfolio's strategies behave in bad months. Every figure is derived further down, with its formula and source. What is missing here is missing on purpose: you will not find a return figure anywhere on this page, because it depends on the market and not on the method. Anyone who shows you one is selling you the market, not their craft.
As of September 25, 2026 · series run dashboard
What is here and what is not
- Process figures: how many strategies were generated, tested and discarded
- Risk figures: depth, duration and shape of the declines, as a percentage of the capital deployed
- How similar the strategies of a portfolio are, especially in its worst months
- What the past does not show, and a measure of how much deeper a decline can still go
- No returns, no account balance, no annualised figures
- No live account: everything is computed on historical price data
- No forecast of what a market will deliver in future
- No outside numbers: everything comes from our own operation, the same one the course comes from
The scale
Everything further down rests on these figures. For each instrument, only the largest pool is counted, not the sum of all computation runs. The reason: a run covering long and short together contains the same strategies as the two separate runs. Adding up all runs gives 6,249 instead of 2,777 strategies and counts the same work several times.
What gets lost along the way
A single run, step by step. The software generates candidates in series. The vast majority fail at the very first condition. What remains goes through the test chain: randomised tests over order, parameters and prices, one and a half times the spread, a retest on real ticks. Right at the end, a few more drop out: if two strategies move almost in lockstep month after month, they count together as one idea. What drops out between the second-to-last and the last stage is decided not by opinion but by a figure with a fixed threshold.
- 1,509,772 candidates generated
- 243 entered the test chain 0.016 % remain
- 38 came through it 15.6 % remain
- 37 independent systems in the end 97.4 % remain
One instrument, one cycle. The same figures as on the home page, told there in context.
One curve, read honestly
Above, the curve of a finished portfolio across the full history, normalised to 100 at the end. So the height tells you nothing about an amount, only about the path. Below, the underwater curve: how far below its previous high the portfolio stood at each point in time, as a percentage of the capital deployed. The lower curve is the more honest of the two: it shows what you have to endure along the way, and the upper one smooths exactly that out in hindsight. Marked are the three deepest declines of the build period and the two validation windows V1 and V2. The portfolio selection did not use these windows, but the test chain of the individual strategies did. That is why they are called validation here, not unseen data.
The chart shows the weekly curve across build period and validation. The figures apply to the build period and are computed on a daily basis. The second figure says how deep it went afterwards in V1 and V2, as a multiple of the deepest decline in the build period. Above 1 means: deeper than anything seen during the build.
The shape of the risk
Each row is one figure across all 36 portfolios. The bar shows the middle half, the line within it the median, the thin ends the outermost cases. A single good number says little; what matters is how tightly or widely the portfolios spread. A tight spread means the method is repeatable, with no single lucky hit standing among many failures.
How far the portfolio fell below its highest level to date in its worst phase, as a percentage of the capital deployed. 19 of the 36 stay under five percent.
Not only how deep, but how long. Deep declines that last a long time weigh more heavily than short ones. The value is the root of the mean of the squared declines.
How much return there is per unit of the deepest decline. The figure says nothing about the size of the return, only about its ratio to the pain endured.
How many days passed until a new high was reached. This is the figure at which most people get out, long before any metric suggests it.
How similar the monthly results of the strategies are over the whole period. Near zero means they do different things.
The same computation, but only over the worst quarter of the months. This is exactly what a portfolio is for. 31 of the 36 sit at zero or below here: their strategies cushion each other instead of falling together.
What share of the depth of a decline falls on the strongest single strategy. Above half means the portfolio is really one strategy with trimmings.
How often several strategies enter at the same hour. High overlap ties up more capital at once than looking at each strategy alone would suggest.
Thin ends: outermost cases. Bar: middle half. Line: median.
What we do not know
Everything above is the past, and the past is a single sample. To estimate what else could have happened, we cut the daily results of each portfolio into blocks of three months and reshuffle them 5,000 times. The blocks stay together so that relationships within a quarter are preserved. The 5,000 paths show how deep the decline goes in the less favourable ones.
The deepest decline you have seen is not the deepest there is.
Blocks of three months, 5,000 draws, fixed random seed so the computation stays repeatable. Plan with the first value, stop at the second.
The check afterwards
The bootstrap estimates from the build period how deep it can get. The two validation windows V1 and V2, each about a year long, show how deep it actually got afterwards. The result is uncomfortable, and that is exactly why it is here.
Deepest decline per window against the deepest decline of the build period, at worst 1.98 times as deep. V1 and V2 are co-optimised: the test chain of the individual strategies saw them, the portfolio selection did not. The real test is still to come: the 36 portfolios have been frozen since September 24, 2026 and are checked against prices from September 1, 2026 onwards. A first evaluation comes in December 2026 at the earliest.
The thresholds
For a figure to mean anything, the point at which it is no longer good enough has to be fixed in advance. Every portfolio has to pass fixed thresholds. They are the same thresholds that stop the runs that never make it onto this page.
If the monthly results of two strategies move in lockstep at 0.7 or tighter, they count as one idea. 2,777 strategies became 1,602 families that way. Without this threshold every portfolio would look broader than it is.
If more than half the depth of a decline falls on a single strategy, the portfolio fails. 29 of the 36 pass that. The remaining 7 still stay in this evaluation. Dropping them would remove them from the spread and make the picture look better than it is.
Limits for decline, standstill and a minimum result per time window are fixed before anything is computed. 36 of the 36 keep to them. A limit you move after seeing the result is not one.
42 instruments
Currencies, commodities and indices. Highlighted are the 15 with finished computed portfolios; for the rest, the test chain is still running, or they were set aside after a measurement run because costs ate up more than the instrument delivered. Setting aside is the more frequent outcome, and that is part of the picture too.
- AUDCAD
- AUDCHF
- AUDJPY
- AUDNZD
- AUDUSD
- AUS200
- CADCHF
- CADJPY
- CHFJPY
- EURAUD
- EURCAD
- EURCHF
- EURGBP
- EURJPY
- EURNZD
- EURUSD
- FCHI40
- GBPAUD
- GBPCAD
- GBPCHF
- GBPJPY
- GBPNZD
- GBPUSD
- GDAXI
- NDX
- NI225
- NZDCAD
- NZDCHF
- NZDJPY
- NZDUSD
- SP500
- SPA35
- STOXX50E
- UK100
- USDCAD
- USDCHF
- USDJPY
- WS30
- XAGUSD
- XAUUSD
- XNGUSD
- XTIUSD
Check the maths
Every figure on this page, with its computation. Each portfolio is computed with capital of one hundred thousand units, however many strategies it holds. All declines are measured against this capital. The curves are normalised to one hundred at the end so they stay comparable and show no amount. All figures apply to the build period; V1 and V2 only appear in the check afterwards.
- Deepest decline
-
max(highest level so far minus level) divided by capitalComputed on a daily basis and over the build period only. The weekly curve that is drawn continues through V1 and V2 and can fall deeper there.
- Ulcer index
-
root of the mean of the squared declines in percentWeights duration as well. Two portfolios with the same deepest decline can lie far apart here.
- Return to decline
-
result divided by deepest declineA ratio, not a size. The numerator appears nowhere on this page, only the ratio.
- Longest standstill
-
largest gap in calendar days between two highsA phase still running up to the last date counts too, otherwise the most recent standstill would look artificially short.
- Correlation
-
mean pairwise correlation of the monthly resultsUnder stress computed over the lowest quarter of the portfolio months, otherwise over all of them.
- Concentration
-
share of the strongest single strategy in the depth of a decline windowIf the share is above half, the portfolio fails.
- Overlap
-
share of entry hours with more than one strategyTogether with the highest number of positions open at once, it forms the measure of the capital required.
- Bootstrap values
-
5,000 draws over blocks of three monthsGiven as a multiple of the historical decline so the figure stays readable without an amount.
These numbers do not come about by themselves.
The method behind them is exactly what is taught in the courses: generate strategies in series, filter them at fixed thresholds, assemble them into portfolios and check their shape before anything runs.